CLIP Colloquium: Shohini Bhattasali (LING/UMIACS)
Zoom: https://umd.zoom.us/j/98806584197?pwd=SXBWOHE1cU9adFFKUmN2UVlwUEJXdz09 (passcode: clip)
Neurocomputational Approaches to Language Processing
Abstract: Utilizing computational approaches to study linguistic processes in the brain has become a recent, emerging area of research. By using tools from natural language processing to operationalize cognitive hypotheses about human language comprehension, we can gain insights into the language and brain connection. This theory-driven experimental approach allows us to use interpretable models against large-scale neuroimaging datasets to test mechanistic hypotheses from cognitive neuroscience that can lead us to future breakthroughs in understanding language processing in the brain and give us clear insights into human cognition.
In this talk, I will share four studies to demonstrate how I have applied various NLP tools to fMRI and MEG datasets to gain insights into several aspects of language comprehension, namely memory retrieval and structure building in language; how argument structure guides incremental sentence processing; comparing broad vs. local contextual prediction in language; parallel processing of local and global linguistic context during speech processing. I will also outline several ways in which I plan to extend this work, especially engaging in more cross-linguistic psycholinguistic and neurolinguistic exploration.